<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>AI Control on Robert's Bytes</title><link>https://pubs.onrender.com/tags/ai-control/</link><description>Recent content in AI Control on Robert's Bytes</description><generator>Hugo</generator><language>en</language><lastBuildDate>Sat, 21 Mar 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://pubs.onrender.com/tags/ai-control/index.xml" rel="self" type="application/rss+xml"/><item><title>Don't build the router yet: a cheap test for monitor ensembles</title><link>https://pubs.onrender.com/posts/monitor_routing/</link><pubDate>Sat, 21 Mar 2026 00:00:00 +0000</pubDate><guid>https://pubs.onrender.com/posts/monitor_routing/</guid><description>AI control teams often propose routing one monitor for subtle attacks and another for obvious ones. Before such a router is built, a three-step diagnostic on existing logs can show whether it can help. On AgentDojo, it cannot.</description></item></channel></rss>